OCZ 512MB EL DDR Dual-Channel Kit
Stephen Cooper, August 4th, 2003
..:: Overclocking Capability ::..
Before we take a look at the performance, let’s take a look at the overclocking capabilities of these OCZ PC4000 sticks. Before I started out going for pure frequency, I decided to lower the CAS Latency to 2.5, while keeping the remaining settings as low as possible, that being 4-4-7. Running at a CAS Latency of 2.5, I was able to take the sticks to roughly 496MHz at the stock voltage of 2.80V. I suspect that if you’re utilizing a motherboard that features advanced VDIMM settings, you’d be able to take this to DDR500 speeds at CL 2.5 even though OCZ only rates the memory for a CAS Latency of 3.0.
Next up, it was time to see how far we could take the memory at 2.80V with stock settings of 3-4-4-8. In our first attempts to overclock the memory past 500MHz, we experienced some serious problems of getting to 500MHz, or past that mark. After updating to the latest BIOS, these problems seemed to disappear, so clearly it was an issue that dealt with the motherboard rather than the OCZ RAM. This has been a problem that has shown up on various motherboards that are currently in the market, no matter what type of PC4000 RAM is being used. Once we were able to take the RAM past the 500MHz mark, we reached a frequency of roughly 524MHz before we started to experience some instability. We upped the voltage to 2.90V, and weren’t able to take the RAM much farther with full stability. After taking a look at our voltage readings, I found that, although the BIOS was set to 2.90V, the actual VDIMM was only 2.84V which explains the lack of a substantial frequency increase. Overall, this is a fairly good overclock, however at this time I am unable to determine if the motherboard is limiting the overclock, or if our beta samples are maxed out.
..:: Performance Benchmarks ::..
As we can see in the SiSoft SANDRA memory bandwidth benchmarks, the bandwidth gains between successive speed increases follow in a nearly linear pattern. At DDR400 settings, we see that the overall memory bandwidth isn’t anything too special for an i875P motherboard, running in the high 4.3GB/s range. As we increase the speed to 433MHz, we see the bandwidth run up to just shy of 4.8GB/s, followed by 5.1GB/s for 466MHz. When we finally attain our 1GHz FSB and DDR500 settings, we see that the memory bandwidth has jumped to roughly 5.5GB/s, and when we take the memory to the maximum stable frequency of 524MHz, we see a bandwidth score of around 5.8GB/s. By upping the FSB and DRAM frequency in sync, we have achieved an extra 1.5GB/s of bandwidth.
Next up for our synthetic performance tests, we have the PCMark2002 results. Much like the results we saw in 3DMark2001 SE, the results we achieve over our test pattern is roughly linear. At 400MHz, we see a PCMark2002 memory result of 7220, while when we up the frequency to 433MHz, and then 466MHz, we see performance marks posted of 7782 and 8455 respectively. When we finally attainted the 500MHz and 524MHz frequencies, we saw our scores shoot up to 8836, and then 9087. This 9087 mark is roughly 25.9% faster than the scores we achieve at stock settings of 400MHz.
Well, now that we’ve seen some numbers thrown at us by the synthetic software benchmarks, let’s see how this additional memory bandwidth can aid our gaming experience. The Q3A benchmarks were run in 640 x 480 mode to avoid any nasty graphics bottlenecks from showing up, however all settings were maxed to make it a little more realistic. When we start things off at 400MHz, we see frame rates of 279.6 for 16-Bit and 278.8 for 32-Bit. As we increase the RAM frequency to 433MHz, we see a good 7.5% jump in frame rates to 302.9, and 299.6 respectively. As we move on up to 466MHz, we see another roughly 7.5% jump in frame rates. The same can be said for the 466MHz to 500MHz frequency jump. Finally, we see our maximum frequency results. With the RAM running at 524MHz, and therefore the FSB running at 1.048GHz, we see a frame rate jump of almost 28% over the stock, 400MHz setting.
Next up for our real world performance tests, we have the usual fan favorite, Unreal Tournament 2003. I used the built-in benchmark that is provided in the latest demo version of UT2003 for all tests, and likewise with the Q3A tests, benchmarks were run in 640 x 480 resolutions. Let’s start off with the FlyBy performance. Here, we see that at stock settings, we see a score of 181.792. When we jump to 433MHz and 466MHz, we see leaps of roughly 8.3% and 7.3% respectively. When we hit the 500MHz mark, we achieve a score of 223.351, while at 524MHz we see a score of 231.291. This is about a 27.2% performance gain over stock settings, very similar to what we saw with the Q3A results. The percentage performance differences we saw in the BotMatch benchmarks all closely math those from the FlyBy benchmarks leaving a final performance lead of roughly 28% for the BotMatch tests.
Previous Page Next Page
|
|
)
)
)
)
)
)